Handle AI detection with caution

ARTIFICIAL intelligence (AI) has created a clear authorship problem. Articles, reviews and literary work can now be produced with little human involvement. It is therefore unsurprising that AI-writing detection programmes have become commonplace. The latest entrant in this field is Pangram Labs, which claims an accuracy of 99.98%.

Pangram is notable because it analyses bodies of public writing and publishes conclusions on whether they are AI-generated. It investigated winners of the Commonwealth Short Story Prize and publicly reported that four of the ten recent regional winners showed strong signs of AI usage in their writing. Pangram also has a browser extension that scans text on feeds such as LinkedIn, Substack and Reddit and inserts labels such as “Human” or “AI” as users browse.

Pangram has its share of critics. The Wall Street Journal editor James Taranto called Pangram a “smear machine” after three editorial columns in his newspaper were labelled as AI-generated. American journalist Taylor Lorenz was accused on X of using AI to write an article for Vanity Fair. Lorenz denied the allegation, and Pangram CEO Max Spero investigated the result and concluded that the detector had made an error.

Today, much of the debate around AI detectors is about their accuracy and the reputational harm to writers. However, for countries that rely on externally developed digital systems, the implications go beyond individual reputation.

Consider this scenario: a respected figure publishes a serious strategic paper that challenges a widely accepted position. It is run through Pangram and tagged as 71% AI-generated. It triggers a social media storm and the writer is discredited. The arguments made in the paper are not even debated. A technical judgment about how a piece of writing may have been produced becomes a verdict about its intellectual credibility. This has deeper implications in an age where authenticity, credibility and belief are increasingly contested.

It could be argued that AI detectors are neutral tools, devoid of any geopolitical significance or power. This would be an extremely simplistic way of thinking. We have been on this ground before. Search engines arrived as simple tools to find what already existed on the Internet. It took many years to realise that Google does not merely retrieve information. It ranks, curates and contextualises it. A change in ranking logic can amplify one political framing and bury another, elevate one interpretation of a crisis, shape what billions encounter first, and therefore what they are most likely to believe.

Social media was initially seen as a neutral connection to interact with friends. We now understand that these platforms shape perceptions, inflame emotions and could even divide societies. Social media has become an indispensable tool in information warfare between nations. In both cases, the consequences went far beyond viewing the technology as an unbiased conduit. AI detection carries a similar potential. It could become another chokepoint through which information is increasingly authenticated and trusted.

Modern societies have gradually outsourced important parts of the information and knowledge environment to technological systems. Search engines, social media platforms, generative AI, and now AI-detection systems are part of an emerging epistemic infrastructure through which societies discover and authenticate knowledge. In this infrastructure, authentication may prove especially consequential because it acts directly on credibility.

A search result can be buried but still found with effort. Dominant narratives on social media can be countered, even if by a minority. However, an important piece of work classified as inauthentic can be discredited before its arguments are even engaged with, making authentication a far more potent gatekeeper of credibility. In a post-truth environment, where trust is already fragile, and competing actors routinely seek to undermine the credibility of institutions and experts, authentication becomes a potential instrument of information control.

For India, this should be viewed as more than an academic integrity issue. We already operate in a vulnerable information environment due to our technological dependencies. India relies extensively on platforms, algorithms and AI systems that are foreign-owned and foreign-governed. These technologies already mediate what information is discovered, amplified and consumed. If externally developed authentication systems also begin determining what is considered genuine, another layer of dependence is added to the information environment.

There is another India-specific issue. Models trained largely on native English set the standard for what human writing should look like, and detectors treat departures from that standard as suspect. A 2023 Stanford study found that widely used detectors flagged genuine essays by non-native English writers as AI-generated more than 60% of the time. Peer-reviewed work this year has reported the same bias. Systematic misclassification of this kind could disproportionately discount Indian scholarship, deepening our reliance on foreign strategic thinking.

This is why the issue goes beyond academic integrity into national security. Nations are already raising security concerns about who holds their data and who controls the networks they depend on. Authentication raises a further question: who decides what is accepted as genuinely human-authored? In a contest increasingly fought over credibility and belief, the ability to certify the authenticity of competing ideas is a strategic capability. For India, which produces much of its intellectual and strategic work in English as a second language, the exposure is sharper.

This does not mean India should reject foreign technologies or attempt to build indigenous substitutes for every detection system. The larger concern is not where Pangram was built, but the authority we choose to give systems like it. If judgments about whether Indian scholarship, journalism or strategic thought are genuinely human come to depend on opaque external systems, a seemingly narrow question of AI detection could become one of intellectual and strategic autonomy.

Lt Gen DS Hooda (retd) is former Northern Army Commander